def main(loader, vocab, opt, model=None):
if model is None:
vocab_size = len(vocab)
model = MultimodalAtt(vocab_size, opt['max_len'], opt['dim_hidden'],
opt['dim_word'])
model = nn.DataParallel(model)
if opt['beam']:
bw = opt['beam_size']
print(f'Using beam search with beam width = {bw}')
model_path = opt['checkpoint_path']
for i in os.listdir(model_path):
if i.endswith('.pth'):
print(i)
path = os.path.join(model_path, i)
model.load_state_dict(torch.load(path))
crit = NLUtils.LanguageModelCriterion()
eval(model, crit, loader, vocab, opt)
else:
'''
Running from inside train.py
'''
crit = NLUtils.LanguageModelCriterion()
scores = eval(model, crit, loader, vocab, opt)
return scores
def main(opt):
dataset = VideoAudioDataset(opt, 'val')
opt['vocab_size'] = dataset.get_vocab_size()
model = MultimodalAtt(opt['vocab_size'], opt['max_len'], opt['dim_hidden'], opt['dim_word'], dim_vid=opt['dim_vid'],
n_layers=opt['num_layers'], rnn_cell=opt['rnn_type'], rnn_dropout_p=opt['rnn_dropout_p']).cuda()
model = nn.DataParallel(model)
crit = NLUtils.LanguageModelCriterion()
for model_path in tqdm(glob.glob(os.path.join(opt['model_directory'],'*.pth'))):
model.load_state_dict(torch.load(model_path))
eval(model, crit, dataset, dataset.get_vocab(), opt, model_path)
def main(opt):
dataset = VideoAudioDataset(opt, 'test')
opt['vocab_size'] = dataset.get_vocab_size()
model = MultimodalAtt(opt['vocab_size'],
opt['max_len'],
opt['dim_hidden'],
opt['dim_word'],
dim_vid=opt['dim_vid'],
n_layers=opt['num_layers'],
rnn_dropout_p=opt['rnn_dropout_p']).cuda()
model = nn.DataParallel(model)
model.load_state_dict(torch.load(opt['model_path']))
crit = NLUtils.LanguageModelCriterion()
eval(model, crit, dataset, dataset.get_vocab(), opt)